Generating conversational gestures from speech audio is challenging due to the inherent one-to-many mapping be-tween and body motions. Conventional CNNs/RNNs assume one-to-one mapping, thus tend predict average of all possible target motions, resulting in plain/boring motions during inference. In order over-come this problem, we propose a novel conditional variational autoencoder (VAE) that explicitly models audio-to-motion by splitting cross-modal latent code into shared motion-specific...

Given that facial features contain a wide range of identification information and cannot be completely represented by single feature, the fusion multiple is particularly significant for achieving robust face recognition performance, especially when there big difference between test sets training sets. This has been proven in both traditional deep learning approaches. In this work, we proposed novel method named C2D-CNN (color 2-dimensional principal component analysis (2DPCA)-convolutional...

Semantic segmentation of high-resolution remote sensing imagery remains challenging due to the coexistence multiscale objects, complex spatial contexts, and ambiguous boundaries. While existing convolutional transformer-based methods have made strides in natural scene understanding, they often underperform on data insufficient detail retention inefficient multi-scale feature modeling. To address these limitations, we propose CDGANet, a novel architecture integrating Cross-Layer Detail-Aware...

Image-based Head Pose Estimation (HPE) from an arbitrary view is still challenging due to the complex imaging conditions as well intrinsic and extrinsic property of faces. Different existing HPE methods combining additional cues or tasks, this paper solves problem by relieving complexity. Our method integrates deep Task-Simplification oriented Image Regularization (TSIR) module with Anchor-Guided (AGPE) module, formulate into a unified end-to-end learning framework. In paper, we define...

Three-dimensional (3D) face models can intrinsically handle large pose recognition problem. In this paper, we propose a novel pose-invariant method via RGB-D images. By employing depth, our is able to self-occlusion and deformation, both of which are challenging problems in two-dimensional (2D) recognition. Texture images the gallery be rendered same view as probe depth. Meanwhile, depth also used for similarity measure frontalization symmetric filling. Finally, texture contribute final...

This paper presents a system that is able to track multiple faces under varying pose (tilted and rotated) reliably in real-time. The consists of two interactive modules. first module performs detection face subject rotations. second does online learning based tracking. A mechanism switching between the modules embedded into automatically decide best strategy for reliable enables smooth transit tracking when one them gives no results or unreliable results. Results demonstrate can make...

In this paper,we used the thermal infrared band of Landsat-5 TM to estimate Land Surface Temperature(LST) with Qin's mono-window algorithm.These results show surface temperature water body and oasis are low;and desert gobi high.These consistent relation hydro-thermal.So algorithm can be spatial distribution land over Zhangye oasis.

In face recognition, searching a person's in the whole picture is generally too time‐consuming to ensure high‐detection accuracy. Objects similar human or multi‐view faces low‐resolution images may result failure of recognition. To alleviate above problems, real‐time recognition method based on pre‐identification and multi‐scale classification proposed this study. The area segmented proportion pedestrian reduce search range, can be robustly detected complicated scenarios such as heads moving...

Face recognition is a very important research topic in computer vision because of its many potential applications. In this paper, we investigated face method based on deep neural network. The sparse coding network and the softmax classifiers were used paper to build train hierarchical after image preprocessing. evaluated ORL, Yale, Yale-B PERET database, respectively. experimental results show that learning can abstractly express original data with efficiency accuracy, achieve good...

Although the face detection problem has been studied for decades, searching tiny faces in whole image is still a challenging task, especially low-resolution images. Traditional methods are based on hand-crafted features, but features of different from those normal-sized faces, and thus robustness cannot be guaranteed. In order to alleviate existing methods, we propose pre-identification mechanism cascaded detector (PMCD) tiny-face detection. This can greatly reduce background other...

The convolutional neural network (CNN) has made certain progress in image processing, language medical information processing and other aspects, there are few relevant researches on its application disease risk prediction. Dyslipidemia is a major modifiable factor for cardiovascular disease, early detection of dyslipidemia intervention can effectively reduce the occurrence diseases. Risk prediction model identify high-risk groups widely used public health clinical medicine. Steel workers...

Electrical energy, as a major source energy of production and life, plays an important role in national economic development. In recent years, the demand electrical is increasing, shortage becomes serious. order to alleviate this problem, people has proposed idea load identification. Load identification system can determine type load, provide detailed electricity consumption running state equipments for supply-side user. This paper gives review monitoring methods loads, compares their...

Passive interferometry technology is based on the relation between reflection and transmission responses of subsurface. The response can be received at surface in presence ambient noise source subsurface with cross-correlation (CC) or multidimensional deconvolution methods. We investigate feasibility electromagnetic (EM) wave passive CC method. design a 2-D finite-difference time domain (FDTD) algorithm to simulate long-duration ground penetrating radar (GPR) measurements random distribution...

Passive time delay estimation in multipath environments is studied this paper. A novel restrained maximum likelihood (ML) estimator proposed to estimate the multiple delays. Unlike traditional ML function which has P global values, restraint conditions limit of paths delays signal with only one value. Markov chain Monte Carlo (MCMC) algorithm used find avoid complex multidimensional grid search, initialization-dependent iterative methods or using interpolation enhance performance. Indeed,...

Cascaded regression has been recently applied to reconstructing 3D faces from single 2D images directly in shape space, and achieved state-of-the-art performance. This paper investigates thoroughly such cascaded based face reconstruction approaches four perspectives that are not well studied yet: (i) The impact of the number landmarks; (ii) vertices; (iii) way using standalone automated landmark detection methods; (iv) convergence property. To answer these questions, a simplified method is...

Reconstructing 3D face models from multiple uncalibrated 2D images is usually done by using a single reference model or some gender/ethnicity-specific models. However, different persons, even those of the same gender ethnicity, have significantly faces in terms their overall appearance, which forms base person recognition faces. Consequently, existing based methods limited capability reconstructing for large variety persons. In this paper, we propose to explore reservoir diverse improve...